A Framework for Rule-Based Management of Parallel Scientific Applications

In this paper we present the design of the rule-based framework for the runtime management of high-performance parallel scientific applications. The framework addresses the formulation of application management behaviors as reaction rules, the injection of rules at runtime, their correct, efficient and scalable parallel enforcement, and the detection and resolution of rule conflicts. Unlike rule-based frameworks in business management and security and resource management domains, the presented framework focuses on high-performance parallel scientific applications, which require consistent and efficient management across processors and components. The framework is part of the Accord programming system (Liu et al., 2004)